🗺️ Architecture Explorer · 8 tracks
The full-system maps: AI Agents, Multi-Agent, MCP, RAG, DeepEval, LLM Evaluations, Prompt Engineering, and Misc AI Tools, each with a playable pipeline.
▸ Explore the tracks🧪 What an LLM really is
Tokens in, tokens out, why raw output drifts, and the guardrails that fix it. Ends at your Local LLM Test Generator.
▸ Play📐 Prompts as power tools
RICE-POT letter by letter: how one master prompt emits an enterprise Selenium or Playwright framework.
▸ Play🗂️ JobTracker AI
A local-first Kanban for the QA job hunt, plus the Easy-Apply agent playbook. Your data never leaves the browser.
▸ Play📋 Chatbot vs agent
What changes when the AI can act: connect Jira, fetch the ticket, draft the full test plan document.
▸ Play🔀 Agents without code
The anatomy of an n8n agent and the 6-flow fleet: chatbot, bug filer, PRD to test cases to Jira or Excel.
▸ Play🌊 Drag-drop agents
LangFlow canvases: a QA Buddy and a bug classifier that assigns category and priority, no code written.
▸ Play⚡ VIBE coding a QA tool
You spec and steer, the AI writes the app: a real FastAPI + React test-case generator built in days.
▸ Play📚 Why RAG exists
The LLM has never seen YOUR product. Chunk, embed, retrieve, cite: from naive RAG to hybrid + reranking.
▸ Play🧭 QA Copilot, end to end
The capstone: one question travels through rewrite, routing, hybrid search, and rerank to a cited answer.
▸ Play🔌 MCP: hands for the AI
The USB-C of AI: how Playwright MCP drives a real browser through your login test and writes the spec.
▸ Play🐍 Python for testers
Twelve lessons from 'what is a variable' to calling an LLM from pure Python. The glue skill for everything else.
▸ Play🧰 Your CSV, now 15 tools
Wrap a 478-row test-case sheet in FastMCP and watch Claude Desktop search, filter, and add cases to it.
▸ Play👥 Three agents, one bug
Triage Analyst, Root Cause Specialist, Test Strategy Advisor: why a crew beats one giant prompt.
▸ Play🏭 Ticket in, code out
Four agents hand work down the line over MCP: analysis, plan, cases, Playwright code, all from one Jira id.
▸ Play🚀 Making it production
Templates, Jira-import CSVs, a generated Playwright framework, and a web UI that fans out over many tickets.
▸ Play⚖️ 'Looks good' isn't a test
Why assertEquals dies on AI output, and the metrics that replace it: relevancy, faithfulness, hallucination.
▸ Play🎛️ Own the model
Prompt, RAG, LoRA, or full fine-tune? The decision path for adapting open models to your private QA data.
▸ Play🧑⚖️ One model judges another
Llama-4 answers, GPT-4.1 grades against a rubric, pytest enforces the threshold. AI quality as a regression suite.
▸ Play📊 The evaluation harness
Three live apps, 29 metric rows, a switchable judge, and a dashboard: the DeepEval capstone.
▸ Play🌐 Selectors break. English doesn't.
A checkout journey written in plain English, executed in a real browser: 22 steps, 118 seconds, passed.
▸ Play🧠 The copilot ramp
From a one-line LLM call to three production agents behind one router: triage, RCA, and test design.
▸ Play